VAST Data's Empire Expands to Google & Microsoft

VAST Data has grown tremendously in the past few years, with a recent growth rate faster than even NVIDIA’s. It’s also associated its name with the biggest neoclouds.

Combining the two, you’d get the impression that VAST is built for the world of massive large language model (LLM) training. You wouldn’t be wrong. But VAST executives believe inferencing is where they will get to show off what the VAST AI Operating System (AI OS) can really do.

HPC and Hyperscale Wins

A few recent announcements exemplify VAST’s level of influence, including work with hyperscalers that points toward enterprise inference usage:

At SC25 this week, VAST announced that the Texas Advanced Computing Center (TACC) will use the VAST AI Operating System (AI OS) as the data layer for its newest supercomputer. Named Horizon, the project will produce a computer 10 times larger than TACC’s current largest academic science high-performance computing (HPC) system, Frontera.

Last week, Google Cloud launched VAST’s AI OS as a fully managed service, with VAST and Google demonstrating that they can connect clusters in near real-time even across continents.

VAST's data services will also be available in Microsoft Azure soon, as the company announced at Microsoft Ignite this week. As with the Google relationship, this will let enterprises use VAST on cloud infrastructure that already holds much of their data and their workloads—and that matters because VAST has its eye on helping enterprises with AI inferencing.

Traction in the Enterprise

AI is forcing enterprises to rethink their relationship to data in general. That’s given VAST an entry point to capture the attention of large enterprises, CEO Renen Hallak said in a recent a call with industry analysts.

That’s a relatively new wrinkle. VAST has long had enterprise clients, but the company has also been pigeonholed as a storage solution, since that’s the fraction of the AI OS that was first available.

Moreover, VAST is a relatively radical storage option, with principles such as a single-tier architecture. There’s no hot data stored close to computing resources, nor cold data in disk drives; all data is in one tier, accessible by any processor any time. Enterprises generally resist tampering with their storage architecture, but the single-tier principle has become necessary as the existing stack proves inadequate for AI performance, Hallak said.

Because VAST represents a system shock to an enterprise’s storage and/or database profile, enterprises want to test the platform out first. Once they’re comfortable, some go all-in with VAST quickly, in some cases deciding to move all their data to the VAST platform. VAST can’t disclose specifics, but some of the customers involved are globe-spanning household names.

Faster than NVIDIA

How is that enterprise traction reflected in VAST’s growth? It’s hard to tell, since VAST has been enjoying massive gains from the frontier AI model builders (likely all the major LLMs, based on Hallak’s comments).

Regardless, between AI builders and enterprises, VAST Data had been tripling its revenue every year coming into 2025, and its growth has accelerated from there. That’s faster growth than NVIDIA—and note that tripling is no small feat, given VAST’s size. It’s a 1,200-employee company that raised $381 million in five rounds.

Still privately held, VAST doesn’t reveal detailed revenues, but executives have said the company has sold more than $2 billion worth of software in nine years. This month, VAST announced a single deal, with neocloud CoreWeave, worth $1.17 billion.

The first $2 billion took nine years, and the second $2 billion is arriving in a heartbeat, as Hallak said.